Kingson Man

Conscious feelings may arise from the mind's reading of the body's condition. We give an artificial agent a homeostatic objective and a body that is vulnerable: its actions have consequences on its ability to continue taking action. When the agent makes decisions by learning to model its own potential futures, its behavior begins to look like its life is at stake. Today's robots and AIs lack feeling because they have no real stake in their own survival. Rewards are extrinsically defined and arbitrary with respect to the agent's own condition and continuation. Homeostatic reinforcement learning begins to bridge this gap by defining reward with respect to the maintenance of the agent's internal states. However, this typically treats the body as a scoreboard, and internal states are not causally linked to the agent's own behavioral competence. A starving agent can move as nimbly as a well-fed one. Here we introduce embodied vulnerability: the agent's internal state directly affects its capacity to act. Allowing the body to stray from its physiological set-point increases the probability of a stumble, or a failure to execute the intended action. Further depletion leads to more stumbles, further impairing one's self-regulation and causing a vulnerability spiral. We train an agent to learn the dynamics of a simple environment and the dynamics of its own body. By simulating potential actions through its inner-world model (an as-if loop), the agent can foresee the consequences to its body and act early to preserve agency. We detect the resulting sparks of artificial feeling, measurable in quality of life and recognizable in quality of behavior.

Kingson Man studies feelings and emotions using artificial systems at Feeling Machines. He earned his PhD in Neuroscience at the University of Southern California, mentored by Antonio Damasio, and was formerly Research Assistant Professor of Psychology. His work has been supported by grants from the National Science Foundation, Google, and private foundations. His research has been covered by the popular press and was featured in Michael Pollan’s bestselling book, A World Appears: A Journey into Consciousness.